On this site
Jev 1.13.0
TypeSafe · System One model
The original, and the model every project on this page is rebuilding.
71.4JevBench v1.7.48 · Open-weights reference row · #15 of 48 on the API board · Capability #12
Jev comparison · JevBench v1.7.48 · updated October 31, 2026
Jev vs OpenJev
One name, ten projects. Here is every one, and how the drop-in compares.
vs
Alternative
OpenJev
razorback16 / Codiv · Jev-compatible server on DiffusionGemma 26B-A4B
Built as a drop-in for TypeSafe’s SDKs, and it can read images.
Not measured in JevBench v1.7.48
- JevHosted API, generally availableWhere it runsOpenJevSelf-host on a 24GB GPU or Apple silicon, or use Codiv’s hosted endpoint
- JevThe original /v1/systemoneRequest formatOpenJevWire-compatible — TypeSafe’s own SDKs work against it unchanged
- JevText onlySupported inputsOpenJevText, plus up to 8 images per request
- JevUp to 255Options per choiceOpenJevUp to 128
- JevCalibrated — 90.6 on JevBenchProbabilitiesOpenJevConfidence derived from output entropy, with no published calibration check
- JevNo thinking modeIn thinking modeOpenJevIntelligence 70.2 in JevBench v1.7.48, above Jev’s 63.6
- JevYes, 71.4 · #15 of 48 on the API boardMeasured in JevBench v1.7.48OpenJevOnly in thinking mode, #72 of 235
Try it on this page
Run Jev 1.14 right here
Edit the example or paste your own text, then run it. Sign in to run: new accounts get 20 welcome credits, and a failed run costs nothing.
Support ticket triage · Jev 1.14
1 Input
Hi, this is the third time I am writing. Our payouts have failed every day since Monday and my team cannot pay suppliers. I already re-entered the bank details twice. If this is not fixed today we will have to move to another provider.
2 Questions
- Choice Which team should handle this ticket?
- Yes / No Does the customer need a response today?
- Score How frustrated is the customer?
- Yes / No Does the customer threaten to leave or switch providers?
Which openJev do you mean?
Every project below has been called openJev or Open-Jev by its author or by someone writing about it. They share a name and an idea, and nothing else — different base models, different sizes, different licences, different quality. Find the one you actually meant before you compare anything.
OpenJevby razorback16 / Codiv
Jev-compatible decision server on DiffusionGemma 26B-A4B (NVFP4, ~18GB). Accepts jev-latest, so TypeSafe SDKs work unchanged. Reads images. Hosted by Codiv, or self-host on a 24GB GPU or Apple silicon.
Default keeps a dated v1.5 score · #72 of 235 at 68.4 in thinking mode
openjev-sglangby ekzhang
Qwen3.6-35B-A3B served on SGLang, also implementing the /v1/systemone wire format. Its documentation states that its probabilities are not calibrated estimates of correctness.
#41 of 48 on the API board at 15.4, measured through its author-hosted endpoint
Open-Jev 9B and 2Bby Zefan Cai
LoRA adapters plus a trained scalar decision head and calibration temperature over a Qwen backbone. Not merged checkpoints — they need the pinned upstream weights and the Open-Jev loader. A 27B v1.1 on Qwen3.8-27B is ranked too.
#102 of 235 at 45.5; 2B #178 at 8.8; 27B v1.1 #115 at 40.5
openJev Verdictby heman10x
A 151M ModernBERT-base encoder with a GLiClass head, post-trained with RLCD and temperature-calibrated. Near-perfect on the data it was fitted to, weak on anything else. Runs on CPU or in a browser tab.
#210 of 235 at 0.2 · Capability #188 at 40.4 (v1.4); other version #216 at 0.1
open-jev-deberta-v3-largeby Kotoba Labs
A DeBERTa-v3 encoder run as a local CPU decision model.
#223 of 235 at 0.0
SemIfby TheoLeeCJ
Was called OpenJev until it was renamed. Reads typed option logits out of a frozen Qwen3.5-4B. Old benchmark tables and blog posts still use the old name, which is where much of the confusion comes from.
#160 of 235 at 16.1 · Capability #143 at 55.4
OpenSourceJevby sabeel111
A Qwen3.5-4B rebuild quantised to Q4_K_M and served with llama.cpp. MIT code over Apache-2.0 weights.
#173 of 235 at 10.6
Open Jev JSON Canvasby JoshuaSP
DiffusionGemma 26B-A4B filling a JSON answer canvas in one denoising step. It returns a label rather than a probability distribution, so it has no calibration score.
#235 of 235 at 0.0 · Capability #204 at 30.6
APUS-OpenJev-v1-9Bby apus-ailab
A merged fine-tune of Qwen3.5-9B, served by its authors’ own native runtime at full depth. The highest-placed project carrying the name after razorback16’s OpenJev in thinking mode.
#90 of 235 at 56.6 · Capability #133 at 56.9
OpenJevby zhangcy122
A separate typed decision API over open LLMs using constrained logprob calibration. Same name as razorback16’s, unrelated codebase.
Not in JevBench v1.7.48
open-jev-typed-decision-engineby eightman999
A 150M typed decision engine that trains on a Colab T4 in about 30 minutes. The author reports 0.697 against Jev’s 0.727 on their own set.
Not in JevBench v1.7.48
open-alternative-jevby IkerMoel
Not an openJev, but constantly mistaken for one: a Qwen3.5-4B rebuild with a similar name and a similar goal.
#175 of 235 at 9.4 · Capability #160 at 49.1
Open Spark Jev (spark-s1)by abhishek085
Not an openJev either, though the name is close: local Qwen3.5-4B decision models built for NVIDIA DGX Spark, Apache-2.0. The measured model is spark-s1-4b-v6.
#98 of 235 at 47.1 · Capability #149 at 54.0
verdict-smallby Manavarya09
Unrelated to openJev Verdict despite the name: a 118M multilingual bi-encoder that scores each option by cosine similarity, runs on CPU and speaks the /v1/systemone wire format.
#225 of 235 at 0.0 · Capability #193 at 37.6
JevBench v1.7.48, measured the same way
One benchmark measured every system under one method — it ranks 235 open-weights systems on one board and 48 hosted API offerings on another, with the same scores in both — so these bars are comparable with each other in a way that vendor-published figures are not. The full board and how to read it are on the JevBench results page.
Composite score
Plain numbers are ranks on the open-weights board; “API” marks a rank on the separate board for hosted API offerings. Jev, a hosted API, is ranked there, and appears on the open-weights board only as an unranked reference row, where its score falls between #49 and #50. The composite hides where systems actually differ, so the charts below break it apart. Benchmark Heaven’s headline Capability ranking, shown next to each score above, averages intelligence and calibration among Jev-class systems.
OpenJev was not measured in JevBench v1.7.48, so there is nothing to chart against Jev here. The analysis below uses the figures its authors publish.
One thing these charts leave out: the composite score above also weighs a fourth axis, cost. We do not reproduce it here; the full table is at the source below.
Every system on the board
The capability scores are charted above; this is the ranking and the deployment detail behind them. Jev and OpenJev are highlighted.
| Model | Rank | Score | Measured latency | Runs on |
|---|---|---|---|---|
| H2O-Lightning-4B v1.1 (H2O.ai, Qwen3.5-4B) | #2 | 77.9 | 0.03s raw, 0.21s adjusted | Self-hosted GPU |
| Xor 26B-A4B (Juspay, Gemma 4 26B-A4B) | #21 | 74.0 | 0.06s raw, 0.27s adjusted | Self-hosted GPU |
| Arcbright Elf 1 31B | #23 | 74.0 | 0.06s raw, 0.27s adjusted | Self-hosted GPU |
| Quyet-1.0-Large (Chinh Nguyen, Gemma 4 31B decoder) | #25 | 73.8 | 0.12s raw, 0.39s adjusted | Self-hosted, H100 |
| Sage 1.3.0 (Levanto Labs, hosted API) | API #11 | 73.7 | 0.17s median / 0.41s p95 | Hosted API |
| decider-12b v2 (Mapika, Gemma 4 12B) | #27 | 73.7 | 0.05s raw, 0.26s adjusted | Self-hosted, RTX PRO 6000 |
| decisio v0.8.0 on Gemma 4 31B (Amin Roudaki, frozen base) | #32 | 73.2 | 0.05s raw, 0.24s adjusted | Self-hosted, H100 |
| decider-12b v1 (Mapika, stock Gemma 4 12B) | #33 | 73.2 | 0.05s raw, 0.26s adjusted | Self-hosted, RTX PRO 6000 |
| Mercury Decide (Inception, diffusion decision API) | API #14 | 72.2 | 0.34s median / 0.96s p95 | Hosted API |
| Winnow-12B Q8 (EldanRing, Gemma 4 12B) | #39 | 72.2 | 0.12s raw, 0.40s adjusted | Self-hosted, RTX 6000 |
| deck-31B (krishna765, frozen Gemma 4 31B) | #45 | 71.9 | 0.13s raw, 0.41s adjusted | Self-hosted GPU |
| Cygnet (blockbrain, frozen Gemma 4 12B) | #46 | 71.9 | 0.04s raw, 0.24s adjusted | Self-hosted, RTX PRO 6000 |
| Jev 1.13.0 (TypeSafe) | API #15 | 71.4 | 0.24s median / 0.31s p95 | Hosted API |
| Jev-Omni (akhilaaa3, Gemma 4 12B) | #52 | 71.0 | 0.16s raw, 0.47s adjusted | Self-hosted, RTX 6000 |
| decisio v0.8.0 on Gemma 4 12B (Amin Roudaki, frozen base) | #54 | 70.6 | 0.12s raw, 0.38s adjusted | Self-hosted GPU |
| torchcast-decision-12b (Torchcast AI, Gemma 4 12B) | #62 | 69.8 | 0.04s raw, 0.24s adjusted | Self-hosted, H100 |
| wity-1 (Wity, closed API, reasoning auto) | API #20 | 69.0 | 1.85s median / 7.97s p95 | Hosted API |
| Plumb-4B (crh225, JevK5 v0.2 + LoRA) | #96 | 50.4 | 0.02s raw, 0.19s adjusted | Self-hosted, H100 |
| decider-4b v2 (Mapika, Qwen3.5-4B) | #111 | 43.1 | 0.03s raw, 0.21s adjusted | Self-hosted, RTX 5090 |
| JevK5 v0.3 (allebee, Qwen3.5-4B) | #119 | 39.2 | 0.02s raw, 0.19s adjusted | Self-hosted, H100 |
| Imajev-4B (Mohit Garg, Qwen3.5-4B LoRA) | #133 | 30.0 | 0.07s raw, 0.29s adjusted | Self-hosted, RTX 5090 |
| Hopper (HopitAI, Qwen3.5-4B LoRA) | #136 | 27.7 | 0.06s raw, 0.26s adjusted | Self-hosted, RTX A6000 |
| SemIf (Qwen3.5-4B) | #160 | 16.1 | 0.06s raw, 0.27s adjusted | Self-hosted, RTX 5090 |
| Laya (ModernBERT-large) | #227 | 0.0 | 0.93s raw, 2.01s adjusted | Self-hosted, CPU |
JevBench method v1.7 applies a 10-second limit per decision: a response taking longer than 10 seconds counts as wrong and contributes zero calibration for that item. Speed uses all decisions, including failures. Benchmark Heaven re-scored recorded per-decision times; it did not run every model again. Tokens, draws and raw outputs are unchanged. Since method v1.8.0 (29 October 2026), self-hosted open-weights rows are costed by GPU time on a monthly-rented RTX PRO 6000 with an efficient inference engine instead of base-model list prices; encoders, architectures outside the six calibrated Gemma/Qwen classes and rows without a verified token basis keep their old costs, and API costs and both caps are unchanged. Intelligence, calibration and latency did not move, but most open-weights cost scores rose, so their composites rose and the Jev-class pool grew. Four rows recorded on a loaded CPU host await re-measurement and remain unranked with their dated v1.6 values; five CPU rows were measured again on GPU and replaced in full. The official composite combines intelligence, calibration, speed and cost using the frozen scorer. Each full-set system answers 1,500 decisions, 300 open and 1,200 sealed; most hosted APIs use an equated 600-item set. Half of intelligence comes from sealed decisions. The open-weights and API boards compare the same measurements within separate hosting groups. Board v1.7.48 ranks 235 open-weights systems and 48 API offerings; an additional API result is preliminary and unranked. There are 110 newer ranked open-weights rows tagged with their draw. Jev 1.13.0 remains an unranked reference on the open-weights board and is ranked on the API board. Capability averages intelligence and calibration among systems inside caps frozen at twice Jev’s v1.5 reference: 217 open-weights systems and 33 APIs qualify. The caps do not track current Jev latency. Self-hosted latency uses an estimated 2× measured latency plus 0.15s; API times include the network. Scores and splits on this site use method v1.8 (v1.7 scoring with the v1.8 cost basis); historical v1.6, v1.5 and v1.4 results are not current standings.
Jev and OpenJev feature by feature
| Attribute | Jev 1.13.0 | OpenJev |
|---|---|---|
| What it is | Hosted System One model from TypeSafe, version jev-1.13.0 | Self-hosted decision server on DiffusionGemma 26B-A4B-it, NVFP4 |
| Licence | Proprietary, hosted | Apache-2.0, weights from NVIDIA and Google, about 18GB |
| Request format | POST /v1/systemone | The same, deliberately: accepts openjev-latest, jev-latest and jev-preview |
| Question types | Noul, choice, score — many per call, answered in parallel | Noul, choice up to 128 options, score over 2–10 ranked levels |
| Inputs | Text only: string, JSON object or array | Text, plus up to 8 images per request |
| Off-schema answers | Typed answers only | Structurally impossible — the model never writes the answer slots |
| Uncertainty handling | Calibrated probability per answer | Re-reads up to 4 times with fresh noise above 0.1 entropy, then averages |
| Hardware | None — it is an API call | 24GB NVIDIA GPU under vLLM, or Apple silicon with 16GB free under MLX |
What each one is better at
Where Jev wins
- Calibration: Jev scores 90.6 in JevBench v1.7.48 against the thinking build’s 81.3, while OpenJev derives confidence from output entropy and publishes no independent calibration check.
- Measured on the current board — 71.4, ranked #15 of 48 on the JevBench v1.7.48 API board — while OpenJev’s default configuration keeps only a dated v1.5 score.
- Up to 255 options per choice against 128, and no incompatibility between images and reasoning, because there are no images.
- Nothing to run: no 24GB GPU, no vLLM, no quantised 18GB checkpoint to keep serving.
- One configuration to reason about. OpenJev\u2019s accuracy, cost and latency all move with denoising steps, sample count and thinking budget.
Where OpenJev wins
- A true drop-in: the same wire format and model aliases, so TypeSafe\u2019s SDKs point at it with a URL change and nothing else.
- Reads images \u2014 up to eight per request. Jev is text only.
- In thinking mode it beats Jev on intelligence in JevBench v1.7.48, 70.2 against 63.6.
- Off-schema answers are structurally impossible, because the model reads the answer slots rather than writing them.
- Apache-2.0, self-hostable on one 24GB GPU or an Apple silicon machine, and also hosted by Codiv.
The analysis
Where Jev and OpenJev really differ, and why the bars above land where they do.
Of everything called openJev, razorback16’s is the one built to replace Jev directly. It is a genuine drop-in: it answers on the /v1/systemone wire format and accepts the model name jev-latest, so TypeSafe’s own SDKs work against it without a code change. JevBench v1.7.48 still lists its default configuration only with a dated v1.5 score; the last release that measured it put it well below Jev on intelligence and calibration.
Its thinking mode was measured, and on the answers it beats Jev: intelligence 70.2 against 63.6, though calibration is 81.3 against Jev’s 90.6. It ranks #72 of 235 open-weights systems at 68.4, below Jev’s 71.4 reference: since Benchmark Heaven’s 29 October GPU-time cost basis its cost score is 54.9, level with Jev’s 54.7, so what holds it back is speed, 72.9, and calibration, and the harmonic mean punishes weak axes hard. If accuracy on genuinely ambiguous inputs is what you are buying, that configuration is more accurate than Jev and you should know it.
What razorback16’s OpenJev actually is
OpenJev, the Jev-compatible decision server on DiffusionGemma, by razorback16 / Codiv.
It runs DiffusionGemma 26B-A4B-it, quantised to NVFP4, about 18GB of weights, under an Apache-2.0 licence from NVIDIA and Google. A diffusion model denoises many tokens at once rather than writing them one after another, and OpenJev turns that into a decision engine: it builds a masked canvas in which only the answer slots are uncertain, then takes one read-only forward pass and reads the probability distribution over the label tokens.
The model never writes into those slots, which is a real structural guarantee rather than a prompt instruction — an answer cannot go off-schema. When the distribution is too uncertain, above 0.1 entropy, it re-reads up to four times with fresh noise and averages the results.
It supports the same three primitives as Jev: noul, choice with up to 128 options, and score over 2 to 10 ranked levels with an expected value. It adds things Jev does not have: up to eight images per request, a tunable 1–8 denoising steps, 1–32 samples per read, and a thinking budget of up to 4,096 tokens.
On hardware, vLLM on a 24GB NVIDIA GPU handles up to 64 in-flight reads at about 94ms p50 at concurrency 1; the MLX backend runs on Apple silicon with 16GB free at 0.2–0.4s on an M3 Ultra, one read at a time. Codiv also hosts it at api.codiv.ai/v1/systemone.
Where its confidence number comes from
OpenJev computes confidence as 1 − H(p)/ln K: one when the model is certain, zero when the distribution is flat. That is a measure of how concentrated the answer is, not a measure of how often it is right. The documentation publishes no independent calibration check, and the last JevBench release to measure it, v1.4.2.2, scored its calibration well below Jev’s.
The practical consequence is the same as with every uncalibrated system on this site: a high number tells you the model was not torn between options, which is not the same as telling you it was correct. Validate the threshold on your own labelled data before you let it auto-approve anything.
The author is straightforward about the rest, too: answer quality is the quality of DiffusionGemma 26B-A4B used this way, choices cap at 128 against Jev’s 255, images cannot be combined with thinking mode, and the Apple silicon backend does not return reasoning content separately.
openJev Verdict: the cautionary one
Verdict is the openJev most likely to mislead you, because its published numbers are excellent and completely honest at the same time. On its own held-out set of 1,000 examples it reports 95.00% top-1 accuracy, a Brier score of 0.0756 and an expected calibration error of 1.13%. In-domain on Banking77 it reaches 95.2%.
On data it was not fitted to, it collapses. Its own author measures 48.07% on TypeSafe’s 337-case benchmark against 90.80% for Jev, and in JevBench v1.7.48 both Verdict builds answer the sealed decisions barely above chance, so their intelligence axis is near zero and they rank #210 and #216 of 235 open-weights systems. Version 1.3.0 of the benchmark, which began measuring intelligence above chance and penalising anything below 50, had already dropped it from #4 to #36 without the model changing at all.
That is not dishonesty and it is barely even a flaw — it is what a 151M encoder is. Verdict was trained on contexts under 71 tokens and runs a 512-token budget with at most 25 candidates. Its README is unusually candid about the failure modes: abstention recall falls from 75.5% to 18.0% against hard-negative siblings, accuracy from 96.0% to 84.75% when distractors are confusable rather than random, and reordering the options flips 4.5% of choices. Fit it to your distribution and it is excellent. Point it at open-ended production text and you get the numbers above.
So which should you use?
Pick Jev if
- You branch on the probability and want it to mean something without fitting it yourself first.
- You would rather not operate a 26B diffusion model, or tune denoising steps and sample counts to trade accuracy against latency.
- Your inputs are text, and you want one stable configuration behind a version you can pin.
Pick OpenJev if
- Your decision involves an image, or you want the option of a reasoning budget on the hard cases.
- You already run GPUs, or the data cannot leave your network, and you want something TypeSafe’s SDKs already speak to.
- You are willing to validate confidence thresholds on your own labelled data rather than trusting them out of the box.
Try Jev on the cases you are actually arguing about
Whichever openJev you landed on, the only comparison that settles it is on your own awkward cases. Running Jev here takes a minute with no setup, and the answers come back with the full probability distribution — which is exactly the thing most of the projects above tell you not to trust yet.
Jev vs OpenJev FAQ
Which project is the real openJev?
None is official. On JevBench v1.7.48, APUS-OpenJev-v1-9B is #90; OpenJev thinking is #72; Zefan Cai’s 9B, 2B and 27B v1.1 are #102, #178 and #115; SemIf is #160; OpenSourceJev is #173; the Verdict builds are #210 and #216; DeBERTa is #223; and JSON Canvas is #235 of 235. openjev-sglang is #41 of 48 on the API board.
Can I point the TypeSafe SDK at OpenJev?
At razorback16’s, yes — it implements the /v1/systemone wire format and accepts jev-latest and jev-preview as model names, so changing the base URL is enough. openjev-sglang does the same; JevBench ranks it on the API board, #41 of 48. The others have their own interfaces.
Is OpenJev as accurate as Jev?
Its default configuration keeps only a dated v1.5 score in JevBench v1.7.48; the last release that measured it put it well below Jev on intelligence and calibration. In thinking mode it is more accurate than Jev, 70.2 against 63.6 on intelligence, but it is slower and less well calibrated, and ranks #72 of 235 open-weights systems at 68.4, below Jev’s 71.4 reference.
Why did openJev Verdict fall from #4 to the bottom of the board?
The model did not change. JevBench v1.3.0 began measuring intelligence above chance, v1.4.0 added a sealed set, and v1.5.0 made the sealed half count for half of intelligence. Verdict barely clears chance on the sealed decisions, so its intelligence axis is near zero and on board v1.7.48 it ranks #214 of 235 open-weights systems at 0.1. The earlier ranking had been rewarding a system that was small and fast but, on this task mix, not far enough above guessing.
What hardware does OpenJev need?
About 18GB of NVFP4 weights: a 24GB NVIDIA GPU under vLLM, which handles up to 64 in-flight reads at roughly 94ms p50, or Apple silicon with 16GB free under MLX at 0.2–0.4s per read on an M3 Ultra. Codiv also hosts it.
Other Jev comparisons
Every number on these pages is quoted from a published source and was read on October 31, 2026.
- Jev vs Imajev-4BA 4B model that reads photos, against the model it copies on text.
- Jev vs decider-4bLevel on speed, far behind on the answers.
- Jev vs JevK5A fast open Qwen3.5-4B system, and far behind Jev on the answers.
- Jev vs CygnetNo training at all, forty-sixth on the open-weights board, and behind Jev on every accuracy axis.
- Jev vs decisioTenth on Capability on its biggest base, and it trains nothing of its own.
- Jev vs metask-jev-rainMore accurate answers than Jev on the benchmark, and two of three rows above it on the composite.
- Jev vs WinnowOne local model for decisions, chat and images, against a hosted decision model.
- Jev vs djevLevel with Jev when it was last measured, and more expensive to run.
- Jev vs SemIfA fast open rebuild on your own GPU, and where it still falls short.
- Jev vs LayaA model you fine-tune on your data, against one that works out of the box.
- Jev vs Mercury DecideTwo hosted decision models that take the same request, one place apart.
- Run Jev locally?The open alternatives in one table, with licences and hardware.
- Jev vs LLMHow a decision model differs from a chat LLM and a BERT classifier.
Sources
- Benchmark Heaven — JevBench board v1.7.48 (open weights)
- Benchmark Heaven — JevBench API board
- razorback16/openjev — Jev-compatible decision server on DiffusionGemma
- openJev Verdict — RLCD ModernBERT decision engine (Apache-2.0)
- Zefan-Cai/Open-Jev — LoRA decision adapters
- SemIf — the project formerly named openjev
- TypeSafe documentation (Jev model card)
OpenJev and every other product named on this page belongs to its respective owner and is not affiliated with Jev AI. Figures are quoted from the sources above as they read on October 31, 2026 and can change.